22 Apr
2016
22 Apr
'16
8:15 p.m.
On Thu, Apr 21, 2016 at 12:07 AM, Raymond Wiker <rwiker at gmail.com> wrote:
I was a bit surprised to see that it used 2901 with a
date code of 1985 -
the 2901 was introduced 10 years before.
The 2901 was the workhorse bit-slice data path chip for many years.
The A, B, and C suffix parts were progressively faster variants
introduced later. Eventually there were CMOS versions, and 16-bit-wide
versions. While AMD introduced the 2903 and 29203 as functionally
improved (but not directly compatible) 4-bit parts, they weren't
nearly as widely used as the 2901.
Most other bit-slice parts can be considered "also-ran" at best, with
the Intel 3001 and 3002 probably being the next most successful. MMI
tried to beat AMD to market with the 5701/6701, which was very similar
to (but not compatible with) the 2901, but they were late to market
and AMD won.
Motorola offered the MC10800 ECL bit slice series, which were
significantly faster at introduction than the contemporary Am2900
parts, but AMD kept introducing faster 2901s. Some later 2901
variants from AMD and National Semiconductor actually used ECL
internally, but had normal TTL I/O, but the CMOS that followed were
even faster than those.
22 Apr
22 Apr
8:25 p.m.
There was a 29G01 offered for a short time. Worth several times their
weight in gold.
--
Will
On Fri, Apr 22, 2016 at 8:15 PM, Eric Smith <spacewar at gmail.com> wrote:
On Thu, Apr 21, 2016 at 12:07 AM, Raymond Wiker
<rwiker at gmail.com> wrote:
I was a bit surprised to see that it used 2901
with a date code of 1985 -
the 2901 was introduced 10 years before.
The 2901 was the workhorse bit-slice data path chip for many years.
The A, B, and C suffix parts were progressively faster variants
introduced later. Eventually there were CMOS versions, and 16-bit-wide
versions. While AMD introduced the 2903 and 29203 as functionally
improved (but not directly compatible) 4-bit parts, they weren't
nearly as widely used as the 2901.
Most other bit-slice parts can be considered "also-ran" at best, with
the Intel 3001 and 3002 probably being the next most successful. MMI
tried to beat AMD to market with the 5701/6701, which was very similar
to (but not compatible with) the 2901, but they were late to market
and AMD won.
Motorola offered the MC10800 ECL bit slice series, which were
significantly faster at introduction than the contemporary Am2900
parts, but AMD kept introducing faster 2901s. Some later 2901
variants from AMD and National Semiconductor actually used ECL
internally, but had normal TTL I/O, but the CMOS that followed were
even faster than those.
8:37 p.m.
On Fri, Apr 22, 2016 at 6:25 PM, William Donzelli <wdonzelli at gmail.com> wrote:
There was a 29G01 offered for a short time. Worth
several times their
weight in gold.
Yes, I forgot about those. Gallium arsenide MESFET for very high
speed. Anyone have data sheets for that family?
8:51 p.m.
I do, back at my old house. I even have a bunch of the old 10G line of
chips, unused. They came out of Collins surplus back in the early
1990s.
--
Will
On Fri, Apr 22, 2016 at 8:37 PM, Eric Smith <spacewar at gmail.com> wrote:
On Fri, Apr 22, 2016 at 6:25 PM, William Donzelli
<wdonzelli at gmail.com> wrote:
There was a 29G01 offered for a short time. Worth
several times their
weight in gold.
Yes, I forgot about those. Gallium arsenide MESFET for very high
speed. Anyone have data sheets for that family? 
10:54 p.m.
Eric Smith mentioned:
[2901 A, B, and C, CMOS versions]
[2903 and 29203]
[Intel 3001 and 3002]
[MMI 5701/6701]
[Motorola MC10800]
I'd add the Texas Instruments SN74S481, SN54LS481 and SN74LS481 TTL 4
bit slices. The Schottky version had a 90ns clock cycle and the low
power versions 120ns. These were 48 pins chips and didn't have an
internal register bank like the 2901. The idea was that you implemented
a memory to memory architecture like the TMS9900.
The TI SBP0400A and SBP0401A were I2L 4 bit slices. The 400 had an
internal pipeline register while the 401 was designed for external
pipelining. ALU operations took 240ns at 200mW. It had an 8 register
bank besides the working register.
All I know about these is what I read in "The Bipolar Micromputer
Components Data Book", December 1977 edition. I have no idea if these
chips were actually shipped or if they were used in any product.
-- Jecel
11:16 p.m.
On 04/22/2016 07:54 PM, Jecel Assumpcao Jr. wrote:
I'd add the Texas Instruments SN74S481, SN54LS481
and SN74LS481 TTL
4 bit slices. The Schottky version had a 90ns clock cycle and the
low power versions 120ns. These were 48 pins chips and didn't have
an internal register bank like the 2901. The idea was that you
implemented a memory to memory architecture like the TMS9900.
Not exactly bit-slice, but how about the National IMP-16 chip set?
--Chuck
23 Apr
23 Apr
12:03 a.m.
On 4/22/2016 9:16 PM, Chuck Guzis wrote:
On 04/22/2016 07:54 PM, Jecel Assumpcao Jr. wrote:
Too Early , Too Slow , Too $$$ is my guess.
With out the 6800/6502 8080/Z80 price wars, how much would a 8 bit
CPU be in the late 70s? $75?
I'd add the Texas Instruments SN74S481,
SN54LS481 and SN74LS481 TTL
4 bit slices. The Schottky version had a 90ns clock cycle and the
low power versions 120ns. These were 48 pins chips and didn't have
an internal register bank like the 2901. The idea was that you
implemented a memory to memory architecture like the TMS9900.
Not exactly bit-slice, but how about the National IMP-16 chip set? --Chuck
Ben.
12:17 a.m.
On 04/22/2016 09:03 PM, ben wrote:
Too Early , Too Slow , Too $$$ is my guess. With out
the 6800/6502
8080/Z80 price wars, how much would a 8 bit CPU be in the late 70s?
$75?
Given the price of memory and other "then-LSI" in the late 70s, $75
doesn't sound unreasonable at all. NSC did bring out the PACE, which
was essentially a single-chip version of an IMP-16 implementation.
But, like all things National, they didn't know how to sell any of their
own stuff.
--Chuck
12:36 a.m.
On 4/22/2016 10:17 PM, Chuck Guzis wrote:
On 04/22/2016 09:03 PM, ben wrote:
What about Radio Shack? $2 for 7400 or 50 cents a gate.
Now the latest INTEL product has how many gates again?
Sadly in hindsight what made the Personal Computer
was the IBM XT. Real Video and Fixed Disk and lots of Memory
and a Warranty. What is on the inside, the greatest profit magin
chips of course.
Ben.
Too Early , Too Slow , Too $$$ is my guess. With
out the 6800/6502
8080/Z80 price wars, how much would a 8 bit CPU be in the late 70s?
$75?
Given the price of memory and other "then-LSI" in the late 70s, $75
doesn't sound unreasonable at all. NSC did bring out the PACE, which
was essentially a single-chip version of an IMP-16 implementation.
But, like all things National, they didn't know how to sell any of their
own stuff.
--Chuck
1:20 a.m.
On 04/22/2016 09:36 PM, ben wrote:
On 4/22/2016 10:17 PM, Chuck Guzis wrote:
What about Radio Shack? $2 for 7400 or 50 cents a
gate. Now the
latest INTEL product has how many gates again?
By the mid 80s, some of us were trying to think of creative uses for the
Z80, which was then going for less than 75 cents in OEM quantities. By
then it was finding its way into PC peripherals.
Time moves on. I'm holding one of those Intel souvenir 80386/80486 key
fobs with the corresponding dice embedded in resin. It really is
startling to see how much larger the 486 die is--and the 486 after
introduction didn't cost proportionately more than the 80386.
--Chuck
1:12 a.m.
On Fri, Apr 22, 2016 at 10:03 PM, ben <bfranchuk at jetnet.ab.ca> wrote:
> Not exactly bit-slice, but how about the National
IMP-16 chip set?
It's bit-slice. The RALU chips were four bits wide, and were used in
at least three different processor architectures, the IMP-4, IMP-8,
and IMP-16. (Despite the similarity of naming, the actual
architectures aren't very similar.)
The CROM (Control ROM) could be custom microcoded for other architectures.
Too Early ,
Not too early to be bit-slice. The DEC PDP-6 was implemented as
bit-slice, but not with ICs.
Too Slow ,
It looks amazingly slow now, but compared to other things available in
1974, it was reasonable.
Too $$$ is my guess.
Actually it was cheap, again by comparison to contemporary things.
With out the 6800/6502 8080/Z80 price wars, how much
would a 8 bit
CPU be in the late 70s? $75?
$395 for an 8080 in 1974. Under $20 for most 8-bitters by the end of 1979.
22 Apr
22 Apr
11:36 p.m.
On 4/22/2016 8:54 PM, Jecel Assumpcao Jr. wrote:
Eric Smith mentioned:
I think the problem was memory at the time. What was the use having
a FAST bitslice machine, but real memory at the time was SMALL and SLOw.
Then if you could get a nice machine, you could never get out of the
School or Lab home with you.
The Lilith computer comes to mind here.
Ben.
PS: http://www.projectoberon.com/ A computer design A few generations
after Lilith. Here a FPGA risc cpu and fast static memory.
[2901 A, B, and C, CMOS versions]
[2903 and 29203]
[Intel 3001 and 3002]
[MMI 5701/6701]
[Motorola MC10800]
I'd add the Texas Instruments SN74S481, SN54LS481 and SN74LS481 TTL 4
bit slices. The Schottky version had a 90ns clock cycle and the low
power versions 120ns. These were 48 pins chips and didn't have an
internal register bank like the 2901. The idea was that you implemented
a memory to memory architecture like the TMS9900.
The TI SBP0400A and SBP0401A were I2L 4 bit slices. The 400 had an
internal pipeline register while the 401 was designed for external
pipelining. ALU operations took 240ns at 200mW. It had an 8 register
bank besides the working register.
All I know about these is what I read in "The Bipolar Micromputer
Components Data Book", December 1977 edition. I have no idea if these
chips were actually shipped or if they were used in any product.
-- Jecel 
23 Apr
23 Apr
12:13 a.m.
On 04/22/2016 10:36 PM, ben wrote:
I think the problem was memory at the time. What was the
use having
a FAST bitslice machine, but real memory at the time was
SMALL and SLOw.
Well, I had 45 ns static RAM for control store on my 2903
bit slice machine. Yes, those were 1K x 4 chips, and I put
heat sinks on them, they ran that hot. Obviously, large
main memory was going to have to run slower.
Jon
12:10 a.m.
I built a 32-bit micro-engine for a project that was
eventually going to be an IBM 360-line CPU.
I pieced the 360, not because it was the greatest design,
but it was VERY well laid-out and would be easy to write
efficient microcode for. I used the 2903 with 2910
controller. I was able to get it to run at 8 MHz, with
3-address operations running at 6 MHz.
But, the project got bogged down, as at a certain point, I
realized HOW MUCH more work lay ahead of me to get a working
system. I had to add 2 more features to the micro-engine -
a 256-way branch from the op-code, and some OR gates to OR
in the register address fields. Then, I had to build a
system bus and memory interface.
(I was going to make the I/O architecture much more like a
PDP-11 than the 360 channel architecture.) Then, I had to
design a general-purpose peripheral controller. I had a
VERY rough sketch for about a 20-chip micro-machine using
(probably) 3 byte-wide EPROMS for instructions) that would
hopefully run at 4 MHz. Then, I had to build a SCSI
controller (I already had a SASI disk on my S-100 system), a
serial mux and a tape controller.
Finally, I had to write at least a primitive OS and figure
out how to come up with compilers for it. Had I known that
UNIX-360 existed, I might have tried to make some kind of
port of that. But, obviously, YEARS of work would have been
needed to make it usable.
See http://pico-systems.com/stories/1982.html for some
pics and description of it.
Jon
11:33 a.m.
On 04/22/2016 11:10 PM, Jon Elson wrote:
Yikes, too many typos, let me try over!
I built a 32-bit micro-engine for a project that was
eventually going to be an IBM 360-like CPU.
I picked the 360, not because it was the greatest design,
but it was VERY well laid-out and would be easy to write
efficient microcode for. I used the 2903 with 2910
controller. I was able to get it to run at 8 MHz, with
3-address operations running at 6 MHz.
But, the project got bogged down, at a certain point, I
realized HOW MUCH more work lay ahead of me to get a
working system. I had to add 2 more features to the
micro-engine - a 256-way branch from the op-code, and some
OR gates to OR in the register address fields. Then, I
had to build a system bus and memory interface.
(I was going to make the I/O architecture much more like a
PDP-11 than the 360 channel architecture.) Then, I had to
design a general-purpose peripheral controller. I had a
VERY rough sketch for about a 20-chip micro-machine using
(probably) 3X byte-wide EPROMS for instructions) that
would hopefully run at 4 MHz. Then, I had to build a SCSI
controller (I already had a SASI disk on my S-100 system),
a serial mux and a tape controller.
Finally, I had to write at least a primitive OS and figure
out how to come up with compilers for it. Had I known
that UNIX-360 existed, I might have tried to make some
kind of port of that. But, obviously, YEARS of work would
have been needed to make it usable.
See http://pico-systems.com/stories/1982.html for some
pics and description of it.
Jon
2:32 a.m.
-----Original Message-----
From: cctalk [mailto:cctalk-bounces at classiccmp.org] On Behalf Of Jecel
Assumpcao Jr.
Sent: Friday, April 22, 2016 10:54 PM
To: General Discussion: On-Topic and Off-Topic Posts
Subject: bit slice chips (was Re: Harris H800 Computer)
Eric Smith mentioned:
[2901 A, B, and C, CMOS versions]
[2903 and 29203]
[Intel 3001 and 3002]
[MMI 5701/6701]
[Motorola MC10800]
I'd add the Texas Instruments SN74S481, SN54LS481 and SN74LS481 TTL 4 bit
slices. The Schottky version had a 90ns clock cycle and the low power
versions 120ns. These were 48 pins chips and didn't have an internal
register bank like the 2901. The idea was that you implemented a memory to
memory architecture like the TMS9900.
The TI SBP0400A and SBP0401A were I2L 4 bit slices. The 400 had an internal
pipeline register while the 401 was designed for external pipelining. ALU
operations took 240ns at 200mW. It had an 8 register bank besides the
working register.
All I know about these is what I read in "The Bipolar Micromputer Components
Data Book", December 1977 edition. I have no idea if these chips were
actually shipped or if they were used in any product.
-- Jecel
-----
Also the later Texas Instruments SN74AS888 (8-bit ALU slice w/ 16x8 register
file and multiply/divide/normalize) & SN74AS890 (sequencer).
Did these ever get any commercial traction?
paul
2:56 a.m.
-----Original Message-----
From: cctalk [mailto:cctalk-bounces at classiccmp.org] On Behalf Of Jecel
Assumpcao Jr.
Sent: Friday, April 22, 2016 10:54 PM
To: General Discussion: On-Topic and Off-Topic Posts
Subject: bit slice chips (was Re: Harris H800 Computer)
Eric Smith mentioned:
I'd say the 74181 (1970) deserves a mention here. Simpler (no register component, ALU
only) but it pretty much kicked off the start of IC-level bit slicing.
Adding to Eric's mention of the PDP-6, the HP2116 (1966) did board-level bit-slicing
(CPU registers, ALU, datapaths on 4 identical boards of 4 bits each).
[2901 A, B, and C, CMOS versions]
[2903 and 29203]
[Intel 3001 and 3002]
[MMI 5701/6701]
[Motorola MC10800]
I'd add the Texas Instruments SN74S481, SN54LS481 and SN74LS481 TTL 4 bit
slices. The Schottky version had a 90ns clock cycle and the low power
versions 120ns. These were 48 pins chips and didn't have an internal
register bank like the 2901. The idea was that you implemented a memory to
memory architecture like the TMS9900.
... 
2:29 p.m.
On 4/22/2016 11:56 PM, Brent Hilpert wrote:
The MC10181 (same thing at the 74181 except implemented in ECL) is used
by the DEC KL-10 in several places.
Rob.
-----Original
Message----- From: cctalk
[mailto:cctalk-bounces at classiccmp.org] On Behalf Of Jecel Assumpcao
Jr. Sent: Friday, April 22, 2016 10:54 PM To: General Discussion:
On-Topic and Off-Topic Posts Subject: bit slice chips (was Re:
Harris H800 Computer)
Eric Smith mentioned:
I'd say the 74181 (1970) deserves a mention here. Simpler (no
register component, ALU only) but it pretty much kicked off the start
of IC-level bit slicing.
Adding to Eric's mention of the PDP-6, the HP2116 (1966) did
board-level bit-slicing (CPU registers, ALU, datapaths on 4 identical
boards of 4 bits each). [2901 A, B, and C, CMOS versions] [2903 and
29203] [Intel 3001
and 3002] [MMI 5701/6701] [Motorola MC10800]
I'd add the Texas Instruments SN74S481, SN54LS481 and SN74LS481 TTL
4 bit slices. The Schottky version had a 90ns clock cycle and the
low power versions 120ns. These were 48 pins chips and didn't have
an internal register bank like the 2901. The idea was that you
implemented a memory to memory architecture like the TMS9900. ...
3008
days inactive
3008
days old
17 comments
9 participants
participants (9)
-
bfranchuk＠jetnet.ab.ca -
cclist＠sydex.com -
elson＠pico-systems.com -
hilpert＠cs.ubc.ca -
jecel＠merlintec.com -
pbirkel＠gmail.com -
radioengr＠gmail.com -
spacewar＠gmail.com -
wdonzelli＠gmail.com